Cat Days of Summer: The Tigers and Schedule Effects
If you’ve been on the internet in the last few weeks (or within earshot of a Michigander) you may have heard about the Tigers. Specifically, you may have heard about how the odds in favor of a Detroit appearance in the 2014 ALDS dropped from 21-to-1 on July 25 to under break-even by August 23 before a slight rebound to finish out the month. Even more specifically, you may have read Mike Petriello’s article about that on this very website. Or at the very least, you may have heard their struggles described in a less quantitative fashion. Regardless, the month of August was not kind to the Bengals.
As Petriello pointed out, this has been less of a Tigers collapse than a Royals surge. But there’s still something to the idea that the Tigers were playing worse in August than they had been previously. Let’s start with the basics:
| 2014 | First Half | August |
|---|---|---|
| R/G | 4.80 | 4.58 |
| RA/G | 4.25 | 4.74 |
| W% | .582 | .516 |
| Pythagenpat | .557 | .484 |
In August, the Tigers scored fewer runs, allowed more runs, and won fewer games than in the first half. On some level, that’s all that really matters. On another level, something else is different about August for these Tigers.
Back on July 14, Buster Olney and Jeff Sullivan both wrote articles about schedule strength. Olney called the Tigers’ schedule the second-most difficult of 17 “contending” teams (paywall), while Sullivan said it was the easiest in all of MLB. One of the key reasons for the discrepancy was that Sullivan was using projections to determine the difficulty of a particular opponent, while Olney was using actual results. Score one for Sullivan. Another key difference was that as of July 14, the Tigers were about to play 55 games in 56 days, which did not factor into Sullivan’s analysis.
A point for Olney? Perhaps. But first, what would we expect to see if this was a result of schedule fatigue? Or put another way, which groups of players might be hurt most or least by not having a day off? Based on conventional wisdom, the bullpen would probably be the most affected, and the starters the least. So how does this match up to the Tigers?
| 2014 | First Half | August |
|---|---|---|
| wRC+ | 115 | 96 |
| SP ERA- | 97 | 102 |
| SP FIP- | 94 | 75 |
| SP xFIP- | 103 | 87 |
| RP ERA- | 108 | 104 |
| RP FIP- | 97 | 117 |
| RP xFIP- | 93 | 122 |
| RP WPA | 0.86 | 1.47 |
This is almost exactly what we might expect, sort of. The relievers lost 29 points of xFIP- from the first half to August, the offense lost 19 points of wRC+, and the starters actually gained 16 points of xFIP-. However, the gap between the starters and relievers virtually disappears when you look at ERA-, and the bullpen WPA is actually higher for August than for the entire first half.
So the picture is perhaps more complicated than conventional wisdom suggests. Regardless, how much of the slump can be directly attributed to the extra games played?
To start to possibly answer this question, let’s look at the Tigers’ performance in 4,777 full games (i.e., with at least 51 recorded outs) since 1984 as a function of the number of games played since the team’s last day off. This doesn’t completely account for the 2014 Tigers’ situation, thanks to double-headers, but it should get us pretty close.
In the two regressions, performance is measured as R or RA in a game above Detroit’s average R/G or RA/G for that season. Opponent strength is measured as average RA/G or R/G above the MLB average R/G for the season, and is included in the regression along with month and home game dummy variables. In both cases, positive coefficients mean more runs, which indicates better performance for the first regression and worse performance for the second. Without further ado, here are the results for the Tigers’ run scoring:
| Estimate | Std. Error | t value | Pr(>|t|) | |
|---|---|---|---|---|
| (Intercept) | -0.310 | 0.137 | -2.263 | 0.024 |
| since_day_off | 0.018 | 0.010 | 1.885 | 0.060 |
| OPP_RA | 0.961 | 0.105 | 9.176 | 0.000 |
| Jul | 0.058 | 0.157 | 0.366 | 0.715 |
| Jun | 0.163 | 0.157 | 1.036 | 0.300 |
| MarApr | 0.052 | 0.168 | 0.308 | 0.758 |
| May | -0.009 | 0.158 | -0.056 | 0.955 |
| SepOct | -0.221 | 0.156 | -1.419 | 0.156 |
| DET_home | 0.166 | 0.093 | 1.795 | 0.073 |
The main effect is nowhere to be found in the runs scored regression, which admittedly still retains a lot of residual variation (R^2 = 0.020). In fact, the coefficient on since_day_off indicates that Tigers batters have historically played better when faced with longer stretches of games. That said, the effect is not significant at the 5% level, and at the 75th percentile of 8 games played without a day off, that’s less than 0.15 extra R/G. Over unusually long stretches, there might be something worth looking at here. However, this not only doesn’t explain the current Tigers’ situation; it actually flies in the face of their August struggles.
So what about pitching? After all, we suggested that the effect of fatigue may be most pronounced for the bullpen.
| Estimate | Std. Error | t value | Pr(>|t|) | |
|---|---|---|---|---|
| (Intercept) | 0.012 | 0.137 | 0.086 | 0.931 |
| since_day_off | 0.005 | 0.010 | 0.546 | 0.585 |
| OPP_R | 0.952 | 0.108 | 8.841 | 0.000 |
| Jul | 0.0104 | 0.156 | 0.067 | 0.947 |
| Jun | -0.106 | 0.156 | -0.683 | 0.495 |
| MarApr | 0.083 | 0.167 | 0.499 | 0.618 |
| May | -0.207 | 0.157 | -1.320 | 0.187 |
| SepOct | -0.129 | 0.155 | -0.833 | 0.405 |
| DET_home | -0.218 | 0.092 | -2.365 | 0.018 |
Just like for runs scored, the runs allowed regression has a very low coefficient of determination (R^2 = 0.018). But this time, the main effect is at least in the right direction, though nowhere near statistical significance. At the in-sample maximum of 38 games the effect is barely 0.2 additional runs allowed per game. In contrast, the average value for since_day_off in August 2014 would be 7.4, which would result in just under 0.04 extra RA/G according to the model. If you treat doubleheaders as canceling the next day off (which was not done in the regression), the expected effect only increases to 0.08 RA/G. Through September 1, the 2014 Tigers allowed 4.39 runs per game on the season, compared to 4.74 in August. Even if you stretch the model in unintended ways, most of their recent slump is still unaccounted for.
Based on this initial analysis, long stretches without days off do not seem to substantially hurt team performance. Further research might arrive at different conclusions, but so far it appears that the 2014 Tigers’ post-break slump is probably just that: a slump. Sometimes, you can’t predict baseball.